Monday, January 6th 2025
Intel Announces Core Ultra 200H Series—Arrow Lake Gets LP Island Cores
The Core Ultra 200H series of mobile processors is designed to cover the majority of mobile device use-cases from the next-generation. These chips are very much based on the latest "Arrow Lake" microarchitecture, but with a few clever design changes. This is a tile-based processor, just like the 200HX series; but the various tiles are different. The Compute tile, which packs the main CPU complex, is noticeably smaller, with 6 "Lion Cove" P-cores, and two "Skymont" E-core clusters with 8 E-cores sharing a 24 MB L3 cache, and a ringbus interconnect. Things get interesting with the SoC tile, which now contains two Low-power Island E-cores. At this point, it is unclear if these are "Skymont," or are older generation "Crestmont" cores, which would mean that Intel has carried over the SoC tile from "Meteor Lake-H."
The SoC tile also contains at 13 TOPS-class NPU, which means these chips miss out on Microsoft Copilot+ AI PC logo. The idea behind this could be that Intel is trying to promote the Core Ultra 200V series "Lunar Lake," which comes with a powerful 45 TOPS NPU. The company is announcing several new models of "Lunar Lake" today, including vPro ones. Switching focus back to "Arrow Lake-H," and our attention is drawn back to the SoC and I/O tiles, which miss out on the number of PCIe lanes, particularly Gen 5 ones, which the Core Ultra 200HX chips come with. The game changer for this chip, however, is the large Graphics tile. The iGPU of "Arrow Lake-H" is based on the same Xe-LPG graphics architecture as the one on "Meteor Lake," and not Xe2 "Battlemage" like on "Lunar Lake." However, this iGPU is vastly different from the one the "Arrow Lake-HX" comes with.
On the "Arrow Lake-HX," the iGPU is based on the Xe-LPG architecture, but only has 4 Xe cores, and each of these Xe cores lacks XMX units—you only get DP4a based acceleration from the execution units. On the "Arrow Lake-H," the iGPU has 8 Xe cores for double the SIMD muscle, and each of these feature XMX units, just like Arc A-series discrete GPUs. The Core Ultra 200H series hence come with AI TOPS from the iGPU ranging between 63 TOPS to 77 TOPS depending on the model, however, Windows 11 will not use the iGPU to accelerate Copilot+ locally. This is more for 3rd party applications that can utilize XMX.
The series is led by the Core Ultra 9 285H, which maxes out the CPU core count of the silicon, with 16 cores on tap—6P+8E+2LP. The P-cores boost up to 5.40 GHz. The iGPU comes with 8 Xe cores. The Core Ultra 7 265H and 255H have essentially the same CPU and iGPU core configurations as the top 285H part, but at lower frequencies. The 265H boosts up to 5.30 GHz, while the 255H only does up to 5.10 GHz on the P-cores.
The Core Ultra 7 235H is where Intel puts out the knife. This chip comes with a CPU configuration of 4P+8E+2LP, with the P-cores boosting up to 5.00 GHz. The iGPU is untouched with 8 Xe cores, though. At the bottom of the lineup is the Core Ultra 5 225H. This one has the same 4P+8E+2LP configuration as the 235H, but also cuts down the iGPU to 7 Xe cores.
The SoC tile also contains at 13 TOPS-class NPU, which means these chips miss out on Microsoft Copilot+ AI PC logo. The idea behind this could be that Intel is trying to promote the Core Ultra 200V series "Lunar Lake," which comes with a powerful 45 TOPS NPU. The company is announcing several new models of "Lunar Lake" today, including vPro ones. Switching focus back to "Arrow Lake-H," and our attention is drawn back to the SoC and I/O tiles, which miss out on the number of PCIe lanes, particularly Gen 5 ones, which the Core Ultra 200HX chips come with. The game changer for this chip, however, is the large Graphics tile. The iGPU of "Arrow Lake-H" is based on the same Xe-LPG graphics architecture as the one on "Meteor Lake," and not Xe2 "Battlemage" like on "Lunar Lake." However, this iGPU is vastly different from the one the "Arrow Lake-HX" comes with.
The series is led by the Core Ultra 9 285H, which maxes out the CPU core count of the silicon, with 16 cores on tap—6P+8E+2LP. The P-cores boost up to 5.40 GHz. The iGPU comes with 8 Xe cores. The Core Ultra 7 265H and 255H have essentially the same CPU and iGPU core configurations as the top 285H part, but at lower frequencies. The 265H boosts up to 5.30 GHz, while the 255H only does up to 5.10 GHz on the P-cores.
The Core Ultra 7 235H is where Intel puts out the knife. This chip comes with a CPU configuration of 4P+8E+2LP, with the P-cores boosting up to 5.00 GHz. The iGPU is untouched with 8 Xe cores, though. At the bottom of the lineup is the Core Ultra 5 225H. This one has the same 4P+8E+2LP configuration as the 235H, but also cuts down the iGPU to 7 Xe cores.
16 Comments on Intel Announces Core Ultra 200H Series—Arrow Lake Gets LP Island Cores
videocardz.com/newz/amd-launches-ryzen-9000hx-fire-range-mobile-cpu-series-up-to-16-zen5-cores-and-140mb-cache
Better yet, a Ryzen AI Max+ 395 with 32 performance threads and 2560 SP cores.
videocardz.com/newz/amd-ryzen-ai-max-300-strix-halo-apus-bring-up-to-16-zen5-cores-and-40-rdna3-5-compute-units
Why is Intel still a thing again?
Anyway, this got old fast:
e.g. no specs, just marketing without details to the cpuflags: www.intel.com/content/www/us/en/products/sku/241747/intel-core-ultra-9-processor-285h-24m-cache-up-to-5-40-ghz/specifications.html
(that is important when you want to optimise software to your hardware)
I doubt any software is ready yet for those dual mixed core thing (p / e cores).
How long will it take to adapt to three different core types with unknown cpuflags.
You probably know that Firerange is using the same chiplet design as the desktop variant wich isn't great for battery life. AMD strix point and Intel H chips are striking a better balance of performance/battery life...and that X3D chip is only going to be found in very expensive beefy laptops with a 5090/5080 :D
I like my 6700K, simple and old, but its replacement isn't going to be Intel - I wonder why?
Lake X vs Y does not. Apparently the SKUs are all out of whack, new xyz can have old abc because reasons. I just can't be bothered investing the time to work it all out, which means writers of news articles that don't spell things out don't get read anymore. Given articles seem to be generated for clicks/reads/etc. and ultimately advertising, then they've lost mine. Assuming I'm not really wierd, then others are like me, hence audience has been narrowed. Unnecessarily I might surmise. YT vids for reviews - I'm getting tech(/marketing wank) overload.You said it in less words than me! Been reading TPU as my goto for as long as I've been a member, but....
Lately on desktop that haven't really been an issue. Skylake was refreshed for a long time but used a new name even though it was the same thing but higher clocked/Higher core count. Then Intel finally decided to call a refresh a refresh, Or AMD would use the + symbol to let people know that it's just a slight optimization rather than a new arch.